Given the prominence and importance of the assessment of likely future demographic trends in Europe, it is surprising to see how little systematic attention the scientific community has been given to the evaluation of arguments underlying the assumptions of future fertility, mortality and migration trends. While the above-described survey shows that the National Statistical Offices put great hope in input from the demographic research community, this topic has largely been confined to the described processes within statistical agencies. Since these offices are in a way forced to make choices on assumptions in order to fulfill their
8 mandate of producing population projections, they cannot escape this challenging task as easily as academics seem to be able to.
The users of population projections – which are by far the most visible and most relevant products that the demographic research community provides to the rest of society – would rightly expect a broadly-based discussion of likely future trends to be the main topic of international population meetings. They would also expect governments and intergovernmental bodies, which greatly rely on the accuracy of population projections in their policy formulations, to commission major studies to make sure that they get the best possible information about likely future demographic trends. But in stark contrast, the reality shows that at scientific meetings in the field of demography as well as in government-sponsored activities around Europe, the discussion of assumptions used in projections is largely absent or at best a marginal topic.
It is important to point out here that the absence of such structured and prominent discussions is not due to the fact that we know all about the future. Quite the contrary, the sense of uncertainty about whether fertility in Europe will recover or continue to decline, or whether we are already close to a maximum life expectancy or will see continued increases, seems to be even higher than in the past. Moreover, studies on the accuracy of past population projections produced since the 1960s have shown that significant errors were made particularly with respect to anticipating the speed of population ageing in Europe. Generally, for most European countries, the national statistical agencies as well as the United Nations Population Division have assumed far too high fertility levels and far too low gains in life expectancy. While these two independent errors tend to cancel out when one is only interested in population size (fewer than expected deaths compensate the effect of fewer than expected births), they strongly reinforce each other when it comes to population ageing (ageing is enhanced by higher life expectancy and by lower fertility). One could even argue that these significant errors of past projections which failed to anticipate the actual speed of ageing have contributed to the fact that today’s societies are not as well prepared for ageing (e.g., in terms of pension systems) as one could have hoped.
Given this situation, the approach presented in this paper will try to show the way for a better inclusion of available scientific knowledge into the process of defining the range of assumptions on future fertility, mortality and migration levels. In other words, it will attempt to facilitate the translation of the vast body of relevant research that exists in the demographic community as well as other related research communities into a definition of specific sets of science-based assumptions for projections. This follows up on earlier work by Lutz, Saariluoma, Sanderson and Scherbov (Lutz et al. 2000), on which the following section partly draws.
As shown in the previous section, up to now this process has mostly happened through the collection of expert opinions. Such procedures typically follow the tradition of Delphi methods that have been well developed and extensively documented elsewhere (Linstone and Turoff 2002). But the problems with expert opinions is precisely that they tend to be opinionated; this can result in all sorts of biases and distortions that are not desirable and do not necessarily reflect the best state of the art in the field. There is abundant evidence that experts tend to hold strong beliefs about the future, which are at the level of emotions and intuitions. Hence, the approach proposed in this paper is nothing short of trying to go beyond opinion-based Delphi and suggest a more objective science-based way. Of course, whenever one has to rely on the views of people in one way or another, this cannot be fully objective, but one can move into this direction by making it inter-subjective and applying the standard scientific tools of peer review and critical evaluation. But in order to make progress in this
9 direction there needs to be something on the table to be evaluated and analytically reviewed.
Hence, the argument-based approach will put specific arguments on the table that are directly relevant for the future course of the demographic force under consideration and which can then be critically assessed.
The following considerations are the product of an interdisciplinary collaboration between demographers involved in population projections and an experimental psychologist working in the field of cognitive science, in a way an expert on experts.
One important contribution that meta-science can make to any scientific approach is to investigate the problems in the way arguments are built in specific scientific fields. Such work can help the applied scientists find a more analytical way of thinking in their own fields. One may ask why is it important to critically inspect the argumentative basis of a science. The logic of the answer is very straightforward: All scientific argumentation ends somewhere and from that point on, the area of intuitive assumptions begins. Infinite chains of arguments are impossible, but we need to be aware of this and reflect on the point when we choose to end the chain of argumentation. This point can be right next to the object of observation, in which case there is no argumentative foundation at all. It can also be too far away from the object in which case the arguments considered and the objects are hardly linked any more. The choice of this cut-off point needs to be based on expert judgment. But this is judgment at the meta-level rather than at the meta-level of the object itself. Such judgment must be based on some sense of plausibility or intuition as it is typically called in cognitive science and foundational analysis.
Intuitions in the foundations of scientific ways of thinking are unavoidable. We cannot get around them; we have to learn how to live with them. The first step in this direction is to understand them in the right manner. The problem with intuitive foundations of science is not that all our intuitions would immediately and necessarily be false, but that we do not know whether they are true and to which degree they are true. This means that we have to adopt a dynamic stance toward them. We have to turn our attention to them and carefully consider the possible strengths and weaknesses in them. When we understand the intuitive foundations better, we are able to use this new understanding for the advancement of science. We can open new perspectives to knowledge and justify the search for new types of knowledge.
Indeed, the ultimate goal of such foundational work is to deepen our understanding of what we are doing. This is a way to speed up the progress in science in general and in the field of making necessary assumptions for population projections in particular.
In the following, we will critically review some of the most common problems with expert judgment and the reference to empirical findings and discuss how arguments should be framed in order to avoid such problems. The goal is not to prove such arguments incorrect or empty in content, but rather to make sure that the specified arguments actually refer to possible causal mechanisms and are specific enough to be falsifiable. Only the evaluation of such arguments will add to our science-based knowledge about likely ranges of future demographic trends.
An important prerequisite for valid argumentation is the clarification of what is the assumed cause and what is the effect. The explanans, i.e., the explanatory premise, refers to statements that explain the explanandum, i.e., the phenomenon which should be explained on the grounds of explanans. In argumentation analysis, it is always central to consider carefully the form and explanatory power of the explanans.
A typical example for the confusion between explanans and explanandum is the assumption that something will not happen because it has not yet been observed, something the literature calls “curve illusion.” In this kind of false argument, one views the shape of an
10 observed curve (the phenomenon to be explained) as the driver that produces a pattern. An example from the field of fertility assumptions is the frequently held view that there is some
“rock bottom” fertility level below which fertility will not fall. This is simply justified by the fact that fertility has never fallen below such a level in any country. There may well be good arguments to assume that fertility will not approach zero in the future, but they cannot be based merely on the description of the “curve” observed so far. Interestingly, in the field of projecting life expectancy, although human history has never experienced a national life expectancy of above the current maximum of some 86 years for Japanese women, few people think that future increases will be impossible. But while a historically unprecedented level is no longer used as an argument in making mortality assumptions, the current practice of (blind) trend extrapolation is not much better from a meta-scientific perspective, if it does not provide any plausible reasons for why life expectancy is assumed to continue to grow at the same speed as in the past or at a decelerating speed, as some agencies assume. More generally, in order to avoid such circularity of taking the explanandum for the explanans, one would have to anchor the argument in the world outside of the curve itself (the observed trends). If such an anchoring is not explicated, the argument cannot be valid.
The circulatory problem also exists when we refer to two different measurements which may be affected by the same cause but do not influence each other. When we measure a fever in a child, we do not think that the high temperature in itself is the illness or that the temperature in the mouth is caused by the temperature in the armpit, even though the correlation would be substantial. Instead, we look for the illness in the body, which explains the high fever measured at both points. We know that the body defends itself from many different types of illnesses by producing a fever and therefore, we look for further symptoms to cancel out incorrect diagnoses and to find the true explanation.
Another problem in this context of defining valid arguments is the confusion of differentials with causes. Much of the social sciences have been inspired by the observation of differentials. Individuals and their behavior differ from place to place, over time and among individuals. These differentials typically give rise to the formulation of explanations as to why the observed patterns of behavior differ. These explanations point the way to the more general causes of behavior. In many cases, however, the analysts stop short of providing real explanations for the observed differentials and suffice by describing only the differentials. An example for such inference from differentials to causes is when people point at the fact that urban women typically have lower fertility than rural women and conclude from this that increasing urbanization will lead to lower fertility. But this conclusion is only correct if it is assured that there is indeed a causal relationship from the kind of living environment on the number of children. There probably are such real causes, but in order to make it a valid argument, the possible causal mechanism has to be identified and discussed. This does not necessarily mean that they have to be proven in the sense of strong causality which may be very difficult. But at least the identified mechanisms should result in a plausible storyline.
Such an argument can then be properly evaluated both with respect to its validity and its relevance.
The same problem affects the currently popular notion of a “second demographic transition” (SDT). It is a name given to a bundle of observed trends in certain values related to sex and partnership and is by its very nature an explanandum (or a “curve” in the above terminology). Although SDT is sometimes referred to as a theory, it does not potentially have predictive power (such as stating that a country that will move into the direction of a more liberal attitude toward sex will have fewer children in the future) and hence must not be mistaken as a testable explanans. In this respect the concept of SDT also suffers from another frequent problem that makes many proposed patterns of explanation inappropriate as valid
11 arguments, namely, the lack of specificity or, in other words, the fact that they are too general.
If an argument is too vaguely formulated or too broad and general in its content so that there is no way to potentially reject it, the argument is not helpful for broadening our science-based understanding of the future.
On the opposite end of the spectrum of problems lie those with arguments that are too specific and too narrow. While such arguments may well be falsifiable in the sense that they have specific information content that can be evaluated in the light of empirical evidence and theoretical cohesion, they may not add much to our overall understanding of likely future trends because they only address a very narrow aspect of all factors that jointly determine the future trend of the demographic force under consideration. A good example for such an argument in the field of fertility determinants is a focus on declining human sperm counts.
While there seems to be convincing evidence that in some countries there have been significant declines in the quantity and quality of sperm counts, and a sufficient number of healthy sperm clearly are a prerequisite for natural conception, some commentators have taken this as an explanation for the declining birth rate. But this relationship is far from straightforward. As discussed in a recent special issue of the International Journal of Andrology (Jørgensen et al. 2006), declining sperm quality and counts may well affect the waiting time to conception (and only in rare cases lead to infertility), but this interacts in a complex way with characteristics of the partner as well as the nature of the partnership.
But the problem of partial explanations is much broader than the example above.
Essentially all arguments about future trends in fertility, mortality and migration focus on certain partial aspects, while leaving others out. Hence, it is one of the most challenging tasks for the development of a new model for argument-based assumption making to bring these different aspects together in a comprehensive way in which the relative importance of the different arguments in determining the future course of the force under consideration are assessed. This will be done in the form of weights to be attached to the different factors that should resemble reality as closely as possible. In other words, we will distinguish between assessing the validity of certain arguments and their relevance in terms of influencing the overall trend of the demographic force.
Based on the above-described considerations and the identification of possible traps and pitfalls in the specification of arguments, in the following section we will present a scheme of core substantive arguments that try to avoid (as far as possible) the above problems. These could become the basis of a systematic future scheme for defining argument-based assumptions for population projections for essentially all countries in the world. While the specific formulation of forces and arguments in the following section is geared toward population projections in industrialized countries, it will be relatively easy to adapt the framework for use in developing countries as well.